OpenAI has launched a new text watermarking system, called textGrain, in ChatGPT and Codex to comply with the European Union’s AI regulations. This rollout initially targets EU users, aiming to enable detection of AI-generated content through embedded, invisible markers.
- Text watermarking deployed first in the European Union.
- API users worldwide can opt in to enable watermarking.
- Detection tools available only to approved researchers initially.
What happened
OpenAI has introduced an invisible, machine-readable text watermarking technology called textGrain in its ChatGPT and Codex models, beginning its rollout exclusively in the European Union. This update reflects efforts to comply with the EU AI Act by enabling detection of AI-generated text outputs. The company claims textGrain performs as well or better than other approaches, such as Google DeepMind’s SynthID, which is also used by Anthropic for watermarking.
The watermarking feature is currently not a global default; rather, OpenAI is focusing on the EU to gather real-world feedback before wider deployment. Additionally, API users worldwide have the option to enable watermarking voluntarily. OpenAI is collaborating with cloud partners to extend this function to model outputs accessed via their platforms in the near future.
Why it matters
This rollout marks a significant step in AI transparency and governance, supporting regulatory compliance with emerging laws such as the EU’s AI Act. Invisible watermarks can help identify AI-generated content without disrupting the user experience, thus assisting platforms, researchers, and policymakers in managing the impact of synthetic text.
However, OpenAI emphasizes that these watermarks do not guarantee detection reliability. They do not verify the content’s accuracy, establish ownership, assess human contribution, or prove human authorship. This highlights the limitations of current watermarking technologies and the need for complementary tools in the fight against misinformation or misuse of AI-generated text.
What to watch next
OpenAI plans to monitor the rollout closely in the EU while collecting feedback to improve watermarking effectiveness and usability. Observers and stakeholders should watch for updates on how the technology performs in real-world applications and whether it will become standard globally.
The availability of detection tools is currently restricted to approved researchers and expert organizations by application, ensuring privacy and minimizing risks of misuse. Future developments might include broader access to these tools, expanded support from cloud providers, and potentially integration with other AI providers’ watermarking solutions to standardize content provenance verification.